Keywords

1 Introduction

In development, tissues and cell types acquire increasing specificity that allows them to fulfill their biological functions. Molecularly, much of this functional specificity is achieved through expression of highly cell-specific gene pathways, coordinated by a multifaceted epigenome. The epigenome here refers to several distinct molecular processes including DNA methylation (DNAm), histones and their post-translational modifications, non-histone chromatin proteins, higher-order chromatin and chromosome organization, and non-coding RNAs. Establishment of these epigenetic features requires precise timing and specificity during the normal development of mammalian cells, throughout every generation. Epigenetic resetting rapidly occurs after fertilization, and then gradual acquisition of epigenetic features occurs in a cell-specific manner, resulting in the variety of cell types that fulfill various biological niches. Most of our knowledge of developmental epigenetics comes from mouse studies, due to the difficulties in collecting and studying human tissues early in development, and it is assumed that similar processes occur in humans. However, important studies verifying (or disputing) these processes in humans are noted. Developmentally significant epigenetic processes are described as “hard-wired” in that they occur without requiring extrinsic signals. However, an understanding of these pre-determined events is essential to also understanding how extrinsic signals, such as from environment, can alter the developmental trajectory of cells into disease-associated states. In this chapter, we provide a foundation for understanding the epigenome in development, and as a consequence, an understanding of the factors that can cause human disease.

2 The Epigenomes of Germ Cells

Germ cell formation involves a series of epigenetic changes including erasure of previous marks, suppression of many genes, activation of genes specific to gametogenesis, and changes in chromatin that facilitate appropriate chromosome interactions during the process of meiosis. As a consequence, genome-wide, sperm and eggs in mammals are highly epigenetically distinct compared to other differentiated somatic cells. Sperm and egg epigenomes are also highly distinct from each other, owing to differences in both the timing and pattern of the epigenomic marks that are established during spermatogenesis and oogenesis.

2.1 Primordial Germ Cells

Primordial germ cells (PGCs) are formed from the epiblast cell layer in the gastrulating embryo through a series of epigenetic remodeling events and expression of specific genes [1]. Chromatin-remodeling events occur as the PGCs differentiate from other epiblast cells, which include structural changes such as heterochromatin decondensation, loss of chromocenters, and increase in nuclear size [2]. Early changes in histone modifications include the loss of H3K9me2, a repressive mark, which is replaced by H3K27me3 from the core component of the polycomb repressive complex 2 [2,3,4]. Repressive methylation is also added onto histone tails of H2A and H4 by nuclear protein complexes, which include Blimp1 and Prtmt5 [5]. Chromatin remodeling continues as the PGCs migrate and take residence in the genital ridges.

Genome-wide erasure of DNAm in PGCs also begins during the migration to the genital ridge. DNA methylation is first lost by an active process where the actions of TET enzymes convert 5’mC to 5’hmC, and then passively lost through successive replication events owing to the absence of normal DNAm maintenance activity (Fig. 5.1) [6, 7]. After migration to the genital ridge and completion of genome-wide demethylation, PGCs achieve a nearly completely demethylated genome [8, 9]. However, a subset of stable marks such as DNAm at imprinted loci and on the inactive X chromosome are removed only after migration to the genital ridge is complete, which may involve additional repression of Dnmt3a, Dnmt3b, and Uhrgf1 [9]. By the end of this process, genomic DNAm drops from around 71–80% to 7–14%, and PGCs enter either spermatogenesis or oogenesis depending on the sex of the embryo [9, 10].

Fig. 5.1
figure 1

Schematic of epigenetic reprogramming during development. The first major genome-wide DNA methylation erasure occurs in primordial germ cells (PGCs), and includes erasure of prior gametic imprints and replacement of H3K9me2 with H3K27me3. H3K27me3 undergoes further reconfiguration during migration to the genital ridge. H3K9me2 is reestablished in growing oocytes and H3K9me3 in growing spermatocytes. Concomitantly, DNA methylation is reestablished, but earlier and to a greater extent in sperm as compared to oocytes (egg). After fertilization, the asymmetric paternal and maternal epigenomes begin reprogramming. H3K4me3, which is distributed in a non-canonical pattern in the growing oocytes, is erased and a non-canonical distribution begins to be established. The paternal genome undergoes active demethylation, and protamines are replaced by histones. The maternal genome also undergoes demethylation, but more gradually through passive replication-dependent mechanisms. Gamete-specific differentially methylated regions (i.e., genomic imprints) and most repetitive sequences are protected during post-fertilization epigenetic reprogramming. At the blastocyst stage, there is the first lineage-specification event of inner cell mass (ICM) and trophectoderm (TE). By the blastocyst stage, most canonical patterns of histone modifications are established, such as bivalent promoters marked by H3K4me3/H3K27me3 at developmental genes. After implantation, DNA methylation is established in both ICM and TE, but in TE remains lower, a difference that is retained through development

Full size image

2.2 Spermatogenesis

In male embryos, sperm progenitors (prospermatogonia) undergo alterations to DNAm starting before birth and continuing until completion of meiotic pachytene after puberty [11, 12]. Sperm DNA becomes highly remethylated owing largely to the activity of DNMT3A and DNMT3L [13]. DNMT3B is involved in methylating imprinted genes and repetitive sequences in sperm, but is inactive in oocytes [14, 15]. PIWI-interacting small RNAs are also responsible for establishing DNAm, which involve Dnmt3c, and H3K9me3 to suppress evolutionarily young copies of retrotransposons [16,17,18]. Another unique aspect of spermatogenesis is that initially, the sperm DNA is tightly wrapped around histones, but the majority are replaced first by non-canonical histone variants, then by transition proteins, and finally by protamines in the haploid stage after maturation, which allows DNA to be tightly packaged inside the sperm head [12, 19, 20]. However, approximately 1% of sperm histones are not replaced by protamines, and are instead retained in the mature sperm DNA specifically at gene promoters [21], suggesting a possible mechanism for transgenerational inheritance of paternal histone marks that can contribute to gene regulation in early development.

2.3 Oogenesis

In female embryos, after PGCs migrate to the genital ridge, there is short period of massive mitotic expansion, after which these oocyte precursors initiate meiosis, but then arrest at prophase 1 and remain dormant in the developing ovary until after puberty. Each month a small subset of this pool of oocytes will undergo a growth phase with only one fully maturing to ovulation. Over their development, oocytes acquire unique DNAm and histone modification profiles. In the late phase of oocyte growth, DNAm increases from 7–14% in PGCs up to 72% in mature oocytes, though remains less than the 86% in mature sperm (Table 5.1) [9, 10]. Unlike somatic cells or spermatocytes, de novo DNAm in oocytes is uncoupled from DNA replication and is acquired in a DNMT3A and DNMT3L dependent manner [22]. In contrast to sperm, where DNAm is more widespread and enriched at enhancers and repetitive short interspersed nuclear elements, DNAm in oocytes is uniquely distributed across actively transcribed gene bodies and CpG islands (CGI) [10], resulting in a oocyte-specific bimodal distribution of hypo- and hyper-methylated regions. Interestingly, DNMT1 and UHRF1, which are canonically responsible for the maintenance of DNAm, function in this context to complete de-novo methylation [15, 23]. Compared to other cell types, oocytes also show increased non-CpG methylation [15]. However, DNAm in oocytes is non-essential for the completion of oogenesis [24, 25]. Lower DNAm in oocytes may allow for subsets of transposable elements to become active in late oocyte maturation, and serve as promoters or first exons for other genes, which can account for ~10% of oocyte-specific transcription [26].

Table 5.1 Major epigenetic differences between tissues and stages in development
Full size table

Oocytes not only have a unique DNAm profile but also display non-canonical distribution of histone modifications. H3K4me3, which is associated with active promoters, is deposited across unusually broad intergenic domains, from the activity of the H3K4 methyltransferase, MLL2 [27,28,29]. H3K4me3 is required for genome-wide transcriptional silencing associated with oocyte maturation and for resumption of meiosis [27, 28, 30]. Although H3K4me3 is anti-correlated with DNAm in mouse oocytes, this inverse relation does not exist in humans [31]. Repressive H3K27me3 is found in large unmethylated regions with low transcriptional activity, but its functional role in oogenesis is unclear. However, a proportion of H3K27me3 in oocytes has been shown to regulate non-canonical DNAm-independent imprinting in the early embryo [32].

During maturation of the oocyte germinal vessicle, chromatin undergoes major conformational changes associated with transcriptional silencing. Upon resumption of meiosis, oocytes lose all higher-order chromatin structures, and interactions become uniform in strength across the chromosome [33, 34].

3 Post-fertilization

After fertilization, the paternal (sperm) and maternal (oocyte) genomes undergo extensive epigenetic reprogramming in order to equalize asymmetric epigenetic differences and prepare the totipotent zygote for cellular differentiation [35]. However, some genomic regions are resistant to this initial reprogramming, leading to the allele-specific expression of genes depending on parental origin, or genomic imprinting.

3.1 Post-fertilization Reprogramming of Gametic Genomes

Initially, the zygote is transcriptionally silent, relying on mostly maternally-inherited proteins and factors to carry out cellular processes during the first initial cell divisions. As maternally-inherited factors are degraded, the transcriptionally silent zygotic genome must become transcriptionally active, and this process, referred to as zygotic genome activation (ZGA), is associated with a variety of epigenetic changes in the paternal and maternal genomes [36]. In the paternal pronucleus, protamines are replaced with histones, and widespread erasure of DNAm is initiated before the first cell division through a rapid active demethylation process controlled by TET1. However, further demethylation occurs through replication-coupled passive dilution, which is aided by TET3 [8, 37]. TET3 converts 5’mC to 5’hmC which then impedes DNAm maintenance [38, 39], and TET3 activity itself can protect against DNMT3a-mediated de novo DNAm [40, 41]. Thus, TET proteins contribute to both active and passive modes of demethylation of the paternal genome after fertilization.

In contrast, the maternal pronucleus appears somewhat resistant to the initial TET demethylation steps, as evidenced by lower TET3 expression and lower TET3-dependent 5’hmC [8]. The resistance of the maternal genome to this initial wave of DNA demethylation is due to H3K9me2-mediated recruitment of PGC7 (aka STELLA or DPPA3), which promotes maintenance of CpG DNAm and inhibits DNMT1-mediated de novo DNAm [42,43,44,45]. The maternal epigenome instead is demethylated by passive dilution through successive rounds of replication. By the blastocyst stage, about 20% of CpGs remain methylated, mostly at imprinted domains and repetitive elements [10, 22, 46].

Histone modifications are also dynamically reconfigured during preimplantation development. H3K4me3 is rapidly depleted after fertilization, but is then replaced by canonical H3K4me3 through H3K4 demethylases KDM5B and KDM1A [27, 28, 47]. In the zygote, there is also erasure of paternally-inherited H3K27me3, but maternally-inherited H3K27me3 at distal intergenic regions is maintained [48]. Despite promoters being devoid of H3K27me3, many still remain transcriptionally inactive, suggesting other silencing mechanisms are at play. H3K27me3 patterning is established by the blastocyst stage, where H3K27me3/H3K4me3 characterizes “bivalent” promoters of developmental genes, which are silent until rapid activation is needed for cell differentiation. Additionally, H3K27me3 maintains expression of some imprinted genes in early development until it is erased in the epiblast lineage at the blastocyst stage, but a subset of H3K27me3 imprinted genes are maintained in the extraembryonic lineage [32]. H3K9 di- and tri-methylation, which are associated to contribute to the silencing and protection of paternal DNAm at imprinted domains, are acquired by the 8-cell stage [43, 49].

From the zygote to blastocyst stages, higher-order chromatin structure such as long-range and local chromosome interactions are gradually established, including more enhancer–promoter interactions, increased number of DNase hypersensitivity sites, and increased chromatin accessibility [33, 34, 50].

3.2 Genomic Imprinting

Genomic imprinting is associated with human developmental disorders such as Prader–Willi Syndrome [51], Angelman Syndrome [52], and Beckwith–Wiedemann Syndrome [53, 54]. Many of the known imprinted genes are regulated by germline differentially methylated regions (gDMRs), where only one parental allele is methylated. As a consequence of differences in epigenetic reprogramming between oogenesis and spermatogenesis, over 1600 CGIs in mouse are gDMRs. Most of these gDMRs are subsequently lost either through removal of DNAm during preimplantation development [22, 55], or through acquisition of de novo DNAm on the alternate allele after implantation [56]. However, some gDMRs are resistant to early epigenetic reprogramming and stable throughout the lifespan, often acting as imprinting control regions (ICRs) that regulate the expression of nearby gene clusters. Mutations within ICRs can lead to loss of the ability to reprogram the appropriate parent-of-origin imprints and can result in unusual inheritance patterns in families. For example, mutations in the ICR regulating the SNRPN gene, implicated in Prader–Willi syndrome, can be passed from mother to child with no direct effect in the first generation. However, when a male inherits an ICR mutation from his mother, the maternal imprint cannot be changed to a paternal imprint in his sperm, thus giving him a 50% risk of having a child affected by Prader–Willi syndrome [57].

In addition to canonical (DNAm-dependent) imprinting, there are a subset of non-canonical imprinted genes for which DNAm is absent, but instead imprinted gene expression is dependent on maternal H3K27me3 [32]. These sites of non-canonical imprinting may also lead to acquired DNAm differences (secondary imprinted DMRs) that are not derived from gDMRs. It is possible that even short-lived imprinting due to H3K27me3 on the maternal genome, leading to paternal-biased gene expression in human preimplantation embryos, could have lasting effects on embryonic development [58, 59]. Further, while the majority of maternal H3K27me3 is lost in embryonic lineages, it appears to be lost more gradually in the placenta and thus allelic differences in gene expression may play a greater role in placental function [12].

In mouse, the greatest numbers of imprinted loci were observed in early embryos, placenta, and brain [60]. A higher rate of imprinted genes in human placenta as compared to somatic tissues is also supported by the large number of placental-specific imprinted DMRs [61]. Placental-specific imprinted DMRs can arise from gDMRs (e.g., imprinted genes Kcnq1 and Igf2r) or through acquisition of secondary DMRs via non-canonical imprinting (e.g., Gab1 and Sfmbt2), with loss of imprinting in the epiblast due to acquisition of DNAm, leading to bi-allelic silencing of the corresponding gene [62]. Interestingly, non-canonical imprinting was localized to endogenous retrovirus-K (ERVK) long terminal repeats, which can act as promoters in the placenta [63, 64]. It is possible that active ERVs in the placenta may interfere with epigenetic silencing in the placental lineages [62]. Placental-specific imprinting appears to be largely species specific and may have no clear function, but some knock-outs of placental-specific imprinted genes in mouse can lead to placental dysfunction [62].

4 Post-implantation

The first lineage-specification event occurs with the differentiation of the trophectoderm (TE) and inner cell mass (ICM) at the blastocyst stage of development. Trophoblast cells, which form both the outer layer of the placental chorionic villi, and the placental cells that invade into the maternal endometrium, are derived from the trophectoderm (TE). The remaining extraembryonic tissues (amnion, placental mesenchyme, and the inner layer of the chorion) derive from the ICM, as do all fetal tissues. At this critical developmental timepoint, there are already extensive genome-wide differences in epigenetic marks between ICM and TE that are essential for regulating the myriad of developmental events that follows.

4.1 Reprogramming of the Blastocyst

The blastocyst has the lowest levels of genome-wide DNAm compared to any other developmental stage, as a result of the genome-wide demethylation that occurs in the first few cell divisions after fertilization. At the blastocyst stage, DNAm begins to be rapidly deposited across all ICM- and TE-derived tissues by DNMT3A and DNMT3B, which preferentially target broadly distributed H3K36me2/3-marked regions [9, 65, 66]. Although DNMT3A and -3B generally do not discriminate between genomic regions, the majority of CGI promoters remain unmethylated to avoid ectopic silencing of important genes. H3K4m3, which is associated with active promoters, protects promoter sequences from de novo DNAm by repelling the ADD domains of DNMT3A and -3B [67]. Developmental gene promoters, which are normally silent but must activate under specific developmental cues, are bivalently marked by PRC2-deposited H3K27me3 and H3K4me2 [68], and are devoid of CGI DNAm owing to TET1 activity [69]. Transcription factors, such as DPPA2, DPPA4, and others, are responsible for the targeted deposition of H3K4me3 and H3K27me3 at CGI promoters of bivalent genes in development [70,71,72].

4.2 Placental Trophoblasts

The embryo develops from the ICM but its growth also depends on the normal development of the placenta, which consists mostly of trophoblasts derived from TE. Due to early divergence of ICM- and TE-derived cell types, the resulting differentiated cell types are each epigenetically unique (Table 5.1). TE is resistant to the dramatic gain of methylation that ICM experiences starting at the blastocyst stage, resulting in a characteristic genome-wide hypomethylated profile of placental trophoblasts that make up the bulk of placenta [73, 74]. Fully developed somatic tissues have a bimodal distribution of genome-wide DNAm, where CGIs are mainly hypomethylated, and the rest of the genome is densely methylated. While unmethylated promoters are conserved in placenta, intergenic regions have lower DNAm compared to somatic tissues, and this lower methylation is organized into large partially methylated domains (PMDs) that are maintained throughout gestation [74, 75]. However, most CpG methylation increases in trophoblast over gestation, even in PMD regions [75, 76]. It is unclear if PMDs regulate transcription directly or are left from earlier transcription regulatory processes. However, even though these regions are relatively gene-poor overall, there is an enrichment for CGI promotors for genes involved in pathways related to the epithelial–mesenchymal transition, immune response, and inflammation [77]. Placental-specific genes and related pathways share epigenetic regulation that is similar to the epigenetic program in some cancers, where trophoblast-specific invasive and immune response pathways are co-opted [77, 78]. A proportion of placental hypomethylation is localized to various repetitive elements, such as LINE-1 [79] and human ERVs [80], which may regulate placental-specific functions [81]. Some repetitive elements serve as alternative promoters for trophoblast-related genes, such as for KCNH5 and IL2RB [80, 82, 83]. Despite observations of PMD- and retrotransposon-driven gene expression in placenta, their contribution to placental function is relatively uncharacterized.

Histone modifications and chromatin dynamics are also under-characterized in placenta. Syncytiotrophoblasts, which line the outer surface of the placental chorionic villi are directly exposed to the maternal environment and govern molecular exchange, express hormones, and produce extracellular vesicles that are critical to regulating interactions between mother and fetus. They are multi-nucleated, created by the fusion of underlying layer of cytotrophoblasts, and their DNA is heterogeneous; older nuclei are condensed into transcriptionally repressed heterochromatin-rich syncytial knots, whereas younger nuclei are euchromatic and transcriptionally active. Recent high-resolution studies indicate histone modifications are important to trophoblast differentiation. Differentiation of syncytiotrophoblast is associated with deacetylation of histone lysine residues, such as depletion of H3K27Ac and H3K9Ac at promoters by histone deacetylases HDAC1 and HDAC2 [84, 85]. Impairment of histone acetylation may be a feature of disease processes in placenta, as aberrant H3K27ac in placenta is associated with maternal preeclampsia and may be a more general feature of placental inflammation [76]. Other histone modifications, such as H3K27me3, H3K4me3, and H3K9me3, are dynamic in cytotrophoblast populations across gestation, but their functional impact on placental function is unclear [76]. H3K9me3, however, is associated with silencing of cytotrophoblast genes, and is correlated with placental DNA hypomethylation [76]. H3K27me3 might be important in establishing a trophoblast-specific transcription program in early placental development, as H3K27me3 has been shown to be uniquely distributed in trophoblast stem cells compared to other embryonic stem cells [86].

The dynamics of epigenetic regulation in the placenta throughout pregnancy is highly specific, especially in comparison to other cell types and tissues, and research in this exciting area is still in its infancy. However, growing interest in the early epigenetic regulation in the placenta promises that future research will lead to exciting new insights into human development.

4.3 X-Chromosome Inactivation

As female 46,XX embryos have two X chromosomes compared to one in 46,XY males, there is initially a sex difference in the expression of X-chromosome genes. As a mechanism for dosage compensation in female embryos, most genes on one of the two X chromosomes are transcriptionally silenced. X-chromosome inactivation (XCI) has been well studied and is a useful paradigm for understanding how monoallelic epigenetic silencing can occur during development. A critical trigger for XCI is the expression of XIST RNA, a 15–17 kb long non-coding RNA that accumulates along one X chromosome in cis and is required for the initiation and stable maintenance of XCI throughout development [87]. While XCI has been most well studied in mouse, there are some key differences with humans. For example, in mouse Xist expression is initiated at the 4–8 cell stage when only the paternal X is inactivated, followed by later reactivation of the paternal X and random XCI in the epiblast. In human embryos, XCI is not initiated until implantation or after, and there is no parental bias in XCI [88]. As human XCI is only completed after implantation and early tissue differentiation, there is a dosage imbalance of most X-linked genes up to this point [89]. Furthermore, as cells can differ by which parental X chromosome is inactivated in females, different cells can express different genetic variants. This not only has consequences for the presentation of genetic disorders but presents challenges to epigenome-wide association studies. Unfortunately, the X-chromosome data is often discarded from epigenome-wide association studies (EWASs) as a consequence.

In human preimplantation embryos, XIST is expressed from both X’s in females, as well as from the single X in males [89]. XACT, a primate-specific X-linked lncRNA, which eventually coats the active X chromosome, is also expressed biallelically at this time [90]. The mechanism by which XIST expression is subsequently dampened to keep only one X active is unclear, although it is hypothesized that XACT may antagonize XIST activity, and XIST may be able to trigger XCI on one randomly chosen X once XACT expression is lost [91]. It is also hypothesized that there is a dosage-sensitive repressor of XIST, encoded by an autosomal gene, that acts prior to the initiation of XCI [88]. In mouse, the RNA binding molecule SPEN has been shown to bind to Xist RNA, and once recruited to the X chromosome, targets active enhancers and promoters and induces repressive chromatin changes that shut down transcription [92]. Regardless of the mechanism, a series of sequential chromatin alterations are established as Xist RNA spreads along the X [93]. First, most euchromatic histone marks are lost, such as promoter/enhancer histone acetylation, and promoter-associated H3K4me3 and H3K36me3. Heterochromatic marks including H3K27me3, H2AK119Ub, and H4K20me1 accumulate later, and there is also accumulation of H3K9me2, and macroH2A. Lastly, CGI promoter methylation secures the stable nature of chromatin compaction characteristic of the inactive. Interestingly, even though H4K20me1 and H3K27me3 are early marks associated with the inactive X chromosome, they are not necessary for silencing and may accumulate as secondary effects of the inactive heterochromatin [94]. These epigenetic changes on the inactive X chromosome result in condensation of the inactive X into a transcriptionally silent structure called a Barr body that localizes to the nuclear periphery.

In addition to higher expression of X-linked genes in females prior to XCI, some genes escape from XCI altogether, and may show persistent higher expression in females compared to males. Roughly 15% of genes in human somatic tissues escape XCI, defined as having expression from the inactive X that is at least 10% of the active X [95]. Genes that escape XCI across different species tend to be enriched for CTCF-binding sites, ATAC-sequencing signal (indicative of open chromatin) and LTR repeats, compared to genes that are subject to silencing by XCI [96]. Thus, some sex differences may be due to persistent gene expression differences on the X, as well as to secondary effects on autosomal gene expression [97]. In addition, some genes show variable XCI, being variably inactivated in some cells, tissues, and individuals. Comparisons across tissues within individuals and in twin pairs show high concordance, indicating that a large portion of variability in XCI escape may be due to cis-acting genetic variation [95, 98].

The process of XCI in the placenta is less well studied and associated with distinct properties. In mouse, there is preferential inactivation of the paternal X, while in human placenta, the process is random but occurs in a patchy fashion across the placenta due to its clonal development [99, 100]. In comparison to males, placental DNAm on the X chromosome in females is greater at gene promoters and lower in gene bodies than expected for X-chromosome inactivation [101]. However, in comparison to blood there is substantial hypomethylation of X-linked promoters in the female placenta, suggesting that there might be more escape from XCI in the placenta [79]. If this is true, then sex differences in fetal development may be driven in part by sex differences in placental function.

5 Factors That Contribute to Variability in Development

Epigenetic mechanisms ensure cellular identities are maintained throughout not only development but the entirety of human life, resulting in wide-ranging types of cellular function that fulfill highly tissue-specific niches. At the same time, extrinsic factors such as environment and aging can perturb the epigenetic program, increasing risk of aberrations that can result in disease. Intrinsic genetic variation is another factor that is increasingly becoming appreciated as a major factor contributing to epigenetic variation, highlighting the importance of considering human diversity.

5.1 Cell-Specific Epigenetics

Fundamentally, epigenetic marks are a cellular process. Despite this, most epigenetic studies rely on whole tissue samples, which are mixtures of distinct cell populations. Because every distinct cell population also has a distinct epigenetic profile, measurements of whole tissues are an average epigenetic profile of the cellular soup that consists of each sample. This results in challenges in interpreting epigenetic changes associated with disease, environment, or development; as observed epigenetic variation can be attributed to either changes that are present in a subset of cells, in all or most cells, or are simply reflecting cell composition variability without any changes in the epigenetic footprint [102]. For example, placental epigenetics, including canonical features like PMDs and placental-specific imprinting, are often features of trophoblast cells, rather than all component cell populations of the placenta [75, 76]. Fortunately, these challenges are surmountable. At the experimental design stage, cell populations most relevant to the research question should be as homogeneous as possible by using cell-specific isolation protocols. When this is not possible, cell composition can be estimated from whole tissue samples, which is commonly done in DNAm studies [103]. Single-cell technologies can address cell-specificity challenges, and have been useful in characterizing early developmental timepoints [104], but current costs and sample preparation requirements remain as major hurdles to employing these technologies at scale on a population level resolution to study disease and environment-related processes.

5.2 Genetic Influences on Epigenetic Variation

Genetic variation at the single nucleotide level is also a major factor contributing to epigenetic variation, especially in contexts where multiple human populations are included. Methylation quantitative trait loci (mQTLs) are where a SNP affects the DNAm of a nearby CpG, which can account for up to 75% of the variance in DNAm [105, 106]. However, it is important to note that joint SNP plus environmental variation are larger contributors to DNAm variation than SNP variation alone [107]. The contribution of genetic and environmental factors can vary depending on tissue and cell type [108], and different human populations [109, 110]. The contribution of genotype-specific DNAm in human development is relatively uncharacterized. Even though the majority of mQTLs in adult brain are also found in fetal brain, a subset are fetal-specific mQTLs and enriched for variants associated with schizophrenia [111] and autism spectrum disorder [112]. Placental-specific mQTLs have also been identified, many of which are associated with expression changes, and a subset overlap known loci linked to fetal growth, diabetes, and inflammation [113]. In a study including blood samples from 4 timepoints, ranging from birth (cord blood) and adulthood (mean age = 47.5), a majority (>84%) of mQTLs were replicated across at all timepoints, suggesting that genetic effects are largely stable across life [114]. These studies highlight the importance of considering the contribution of human diversity and genetic influences on epigenetic processes in human development.

6 Transgenerational Epigenetics

It has been demonstrated in many organisms that a subset of environmentally induced epigenetic changes may be transmitted through the germline over multiple generations [115]. For example, exposures to endocrine disrupting chemicals, altered nutrition (e.g., high fat or low calorie), and stress have all been associated with transgenerational epigenetic alterations in rodent models [116]. In mammals, there is more evidence for male transgenerational epigenetics, as it is difficult to separate the in utero effects of maternal or grandmaternal exposures from true transgenerational inheritance in females [115, 116]. Epigenetic inheritance may potentially involve DNAm, histone modifications, short or long non-coding RNAs, and structural chromatin proteins [117, 118]. Although germline remodeling in spermatogenesis involves DNAm erasure and replacement of most histones with protamines, a small number of histones still remain, particularly in active CpG-rich gene regulatory regions [119, 120]. In a transgenic mouse model with overexpression of the histone demethylase KDM1A leading to depletion of H3K4me2, there was enrichment for H3K4me3 that escaped reprogramming in the embryo, supporting that some histone marks can serve as a potential mechanism for transgenerational inheritance [121]. Other studies have supported a role for small non-coding RNAs in transgenerational epigenetic inheritance. For example, male mice subjected to early life stress showed altered stress response pathways in their offspring, and by isolating microRNAs from the sperm of the stress-exposed males and injecting them into zygotes, similar behavioral effects were reproduced in offspring from control (non-stressed) parents [122, 123]. In another model whereby rats were transiently exposed to agricultural fungicide vinclozolin or pesticide DDT, it was observed that changes in DNAm, histone localization, and non-coding RNAs colocalized to the same chromosomal regions, implying integrated effects that are mediated by non-coding RNA directed DNAm and DNAm-directed histone retention [124].

The possibility of epigenetic inheritance in humans is suggested by the Dutch famine, whereby the offspring of males, but not of females, prenatally exposed to caloric restriction showed increased adiposity [125]. Furthermore, small non-coding RNA expression and DNAm differed in the sperm from obese as compared to lean men [126]. Altered microRNA profiles were also reported in the sperm of men who smoked [127], while altered DNAm was reported in the sperm of infertile men taking high dose folate supplements [128]. However, direct evidence for epigenetic inheritance in humans is scarce, and in many cases where abnormal DNAm appeared to be transmitted in families, were actually due to genetic mutations that influence DNAm (e.g., imprinting mutations) [117, 129]. Likely, the same mechanisms shown in other mammals occur also in humans, but their significance in the context of the extensive genetic, environmental, and cultural variation in humans remains to be determined.

7 Conclusion

In conclusion, an important area of human epigenetics is understanding the epigenetic processes that occur during early development and gametogenesis. Histone modifications, chromosome structure, DNAm, and non-coding RNAs are highly dynamic in development, and are essential in setting up the epigenetic profiles that regulate gene expression in differentiated cells in later life. Recent and ongoing advancements in areas such as single-cell and low-input sequencing technologies are rapidly enabling characterization of the multifaceted epigenome at critical developmental timepoints. An important area for future research will be to not only continue measuring these epigenetic marks with advancing technologies, but also to understand the functional relevance of various epigenetic processes, how they contribute to development itself, and what factors disturb these processes that results in increased risk of disease in normal development. Recently developed ex vivo organoid technologies are promising, as more mechanistic studies and particularly human ones are needed. Although major epigenetic processes are often conserved across mammals, there is often evolutionary divergence in the timing and function of the proteins involved. Population studies measuring epigenetic marks in early life tissues, will also be important to understand how variability in normal development occurs under the context of genetic and environmental-related variation. Technology advancements and growing interest promise that the dynamics of the epigenome in development will be characterized at an unprecedented level of detail than ever before. This will lead to many exciting new insights that will aid our understanding of how genetic and environmental factors interact with the epigenome to create the vast diversity of human phenotypes.